LLMs aren’t the only type of AI, so why are they so popular? These are my thoughts.

Language is the way we interact as humans—gestures, sound, and especially symbols or text. It’s how we interact with the world around us.

Let's say you want a coffee from Starbucks, you'd walk up and talk to the barista. Now they will go and get your order having understood what you said and what you meant. Based on the barista's understanding of what you want, they know all the things to do to get your desired beverage. Imagine having a personal butler or assistant you can interact with in this way to get stuff done. This pattern of interacting with a versatile assistant in natural language is key.

LLMs have taken an important seat in the AI space today because they have become the most common way we interact with AI. They are peripheral systems, the frontend—the user interface—to a great number of AI and non-AI services. We call these services tools.

Tool calling enables LLMs to make API calls, run custom code, or even call other AI models. Just like our Starbucks barista, the model talks to you to know what you want, then works behind the scenes using tools to achieve the goal. As you see by now, it's not just text generation. LLMs are trained on text that includes prompts like “generate an image of…” paired with corresponding actions or outputs (e.g., calling an image model). They learn associations between certain phrases and actions. LLMs can therefore map intent to tool calls.

AI you can talk to is what appeals to the masses. LLMs will only get better and more powerful with time. I'm excited to see what humans create next.